Hybrid search
Search strategy that combines dense (vector) and sparse (BM25/lexical) — captures both semantic similarity and exact term matching.
Dense is good at paraphrases and conceptual similarity ("car" ↔ "automobile") but can miss specific tokens (product IDs, rare names). BM25 finds exact terms but doesn't understand synonyms. Hybrid gets both.
Fusion is usually done with Reciprocal Rank Fusion (RRF): sum 1/(k + rank) across the two lists, sort. Supported natively by Elastic, Weaviate, Qdrant, OpenSearch, pgvector + tsvector. Default in modern RAG pipelines.